Update app.py

app.py

@@ -135,7 +135,7 @@ def prepare_output_dir(input_file, output_dir):
         raise
     return out_dir
 
-def roformer_separator(audio, model_key, seg_size, overlap, model_dir, out_dir, out_format, norm_thresh, amp_thresh, progress=gr.Progress()):
+def roformer_separator(audio, model_key, seg_size, override_seg_size, overlap, pitch_shift, model_dir, out_dir, out_format, norm_thresh, amp_thresh, progress=gr.Progress()):
     """Separate audio using Roformer model."""
     print_message(audio, model_key)
     model = ROFORMER_MODELS[model_key]
@@ -151,7 +151,9 @@ def roformer_separator(audio, model_key, seg_size, overlap, model_dir, out_dir,
             mdxc_params={
                 "batch_size": 1,
                 "segment_size": seg_size,
+                "override_model_segment_size": False,
                 "overlap": overlap,
+                "pitch_shift": pitch_shift,
             }
         )
 
@@ -169,7 +171,7 @@ def roformer_separator(audio, model_key, seg_size, overlap, model_dir, out_dir,
         logging.error(f"Roformer separation failed: {e}")
         return None, None
 
-def mdx23c_separator(audio, model, seg_size, overlap, model_dir, out_dir, out_format, norm_thresh, amp_thresh, progress=gr.Progress()):
+def mdx23c_separator(audio, model, seg_size, override_seg_size, overlap, pitch_shift, model_dir, out_dir, out_format, norm_thresh, amp_thresh, progress=gr.Progress()):
     """Separate audio using MDX23C model."""
     print_message(audio, model)
     try:
@@ -184,7 +186,9 @@ def mdx23c_separator(audio, model, seg_size, overlap, model_dir, out_dir, out_fo
             mdxc_params={
                 "batch_size": 1,
                 "segment_size": seg_size,
+                "override_model_segment_size": override_seg_size,
                 "overlap": overlap,
+                "pitch_shift": pitch_shift,
             }
         )
 
@@ -336,7 +340,9 @@ with gr.Blocks(
                 roformer_model = gr.Dropdown(label="Select the Model", choices=list(ROFORMER_MODELS.keys()))
             with gr.Row():
                 roformer_seg_size = gr.Slider(minimum=32, maximum=4000, step=32, value=256, label="Segment Size", info="Larger consumes more resources, but may give better results.")
+                roformer_override_seg_size = gr.Checkbox(value=False, label="Override segment size", info="Override model default segment size instead of using the model default value.")
                 roformer_overlap = gr.Slider(minimum=2, maximum=10, step=1, value=8, label="Overlap", info="Amount of overlap between prediction windows.")
+                roformer_pitch_shift = gr.Slider(minimum=-12, maximum=12, step=1, value=0, label="Pitch shift", info="Shift audio pitch by a number of semitones while processing. may improve output for deep/high vocals.")
         with gr.Row():
             roformer_audio = gr.Audio(label="Input Audio", type="filepath")
         with gr.Row():
@@ -351,7 +357,9 @@ with gr.Blocks(
                 mdx23c_model = gr.Dropdown(label="Select the Model", choices=MDX23C_MODELS)
             with gr.Row():
                 mdx23c_seg_size = gr.Slider(minimum=32, maximum=4000, step=32, value=256, label="Segment Size", info="Larger consumes more resources, but may give better results.")
+                mdx23c_override_seg_size = gr.Checkbox(value=False, label="Override segment size", info="Override model default segment size instead of using the model default value.")
                 mdx23c_overlap = gr.Slider(minimum=2, maximum=50, step=1, value=8, label="Overlap", info="Amount of overlap between prediction windows.")
+                mdx23c_pitch_shift = gr.Slider(minimum=-12, maximum=12, step=1, value=0, label="Pitch shift", info="Shift audio pitch by a number of semitones while processing. may improve output for deep/high vocals.")
         with gr.Row():
             mdx23c_audio = gr.Audio(label="Input Audio", type="filepath")
         with gr.Row():
@@ -365,10 +373,10 @@ with gr.Blocks(
             with gr.Row():
                 mdx_model = gr.Dropdown(label="Select the Model", choices=MDXNET_MODELS)
             with gr.Row():
-                mdx_hop_length = gr.Slider(minimum=0.001, maximum=0.999, step=0.001, value=0.25, label="Hop Length")
+                mdx_hop_length = gr.Slider(minimum=32, maximum=2048, step=32, value=1024, label="Hop Length")
                 mdx_seg_size = gr.Slider(minimum=32, maximum=4000, step=32, value=256, label="Segment Size", info="Larger consumes more resources, but may give better results.")
                 mdx_overlap = gr.Slider(minimum=0.001, maximum=0.999, step=0.001, value=0.25, label="Overlap")
-                mdx_denoise = gr.Checkbox(value=True, label="Denoise", info="Enable denoising during separation.")
+                mdx_denoise = gr.Checkbox(value=False, label="Denoise", info="Enable denoising during separation.")
         with gr.Row():
             mdx_audio = gr.Audio(label="Input Audio", type="filepath")
         with gr.Row():
@@ -384,8 +392,8 @@ with gr.Blocks(
             with gr.Row():
                 vr_window_size = gr.Slider(minimum=320, maximum=1024, step=32, value=512, label="Window Size")
                 vr_aggression = gr.Slider(minimum=1, maximum=50, step=1, value=5, label="Agression", info="Intensity of primary stem extraction.")
-                vr_tta = gr.Checkbox(value=True, label="TTA", info="Enable Test-Time-Augmentation; slow but improves quality.")
-                vr_post_process = gr.Checkbox(value=True, label="Post Process", info="Enable post-processing.")
+                vr_tta = gr.Checkbox(value=False, label="TTA", info="Enable Test-Time-Augmentation; slow but improves quality.")
+                vr_post_process = gr.Checkbox(value=False, label="Post Process", info="Enable post-processing.")
                 vr_post_process_threshold = gr.Slider(minimum=0.1, maximum=0.3, step=0.1, value=0.2, label="Post Process Threshold", info="Threshold for post-processing.")
                 vr_high_end_process = gr.Checkbox(value=False, label="High End Process", info="Mirror the missing frequency range of the output.")
         with gr.Row():
@@ -401,7 +409,7 @@ with gr.Blocks(
             with gr.Row():
                 demucs_model = gr.Dropdown(label="Select the Model", choices=DEMUCS_MODELS)
             with gr.Row():
-                demucs_seg_size = gr.Slider(minimum=1, maximum=100, step=1, value=50, label="Segment Size")
+                demucs_seg_size = gr.Slider(minimum=1, maximum=100, step=1, value=40, label="Segment Size")
                 demucs_shifts = gr.Slider(minimum=0, maximum=20, step=1, value=2, label="Shifts", info="Number of predictions with random shifts, higher = slower but better quality.")
                 demucs_overlap = gr.Slider(minimum=0.001, maximum=0.999, step=0.001, value=0.25, label="Overlap")
                 demucs_segments_enabled = gr.Checkbox(value=True, label="Segment-wise processing")
@@ -422,7 +430,9 @@ with gr.Blocks(
             roformer_audio,
             roformer_model,
             roformer_seg_size,
+            roformer_override_seg_size,
             roformer_overlap,
+            roformer_pitch_shift,
             model_file_dir,
             output_dir,
             output_format,
@@ -437,7 +447,9 @@ with gr.Blocks(
             mdx23c_audio,
             mdx23c_model,
             mdx23c_seg_size,
+            mdx23c_override_seg_size,
             mdx23c_overlap,
+            mdx23c_pitch_shift,
             model_file_dir,
             output_dir,
             output_format,